Cloud.ru открыл исходный код Guardrails Filter для защиты данных в запросах к LLM

In July 2026, Cloud.ru open-sourced GuardrailsFilter — a transparent reverse proxy that strips sensitive data from requests to language models and restores it in the responses. The tool is published as open source and available for free.

What GuardrailsFilter Does

GuardrailsFilter works as a transparent reverse proxy between the client application and the LLM provider: it intercepts the request before it goes to the model, finds sensitive data in it and replaces it, and in the model's response performs the reverse operation — returning the original values to their place. For the application and for the model itself this happens invisibly, and there is no need to rewrite integration code.

The key point here is that data does not leave the perimeter in the open. The model sees anonymized text, while the user receives a normal response with the values restored.

• Developer — Russian cloud provider Cloud.ru
• Type — transparent reverse proxy
• Function — removing sensitive data from the request and restoring it in the response
• Compatibility — works with any LLM providers
• License — open-source, distributed free of charge

Why This Matters for Business

The problem GuardrailsFilter solves is familiar to any company connecting cloud LLMs to its workflows: prompts inevitably end up containing customers' personal data, contract numbers, internal documents. All of this goes to the model provider's side, and the sender can no longer control what happens to the data afterward.

The proxy breaks this dependency. A layer appears between the application and the model that guarantees only anonymized text leaves the perimeter. This removes part of the risk under personal data protection requirements (152-FZ) and corporate security policies, and makes it possible to use even external models in places where the security team previously blocked any data transfer.

Why the Cloud Provider Is Giving Away the Code for Free

GuardrailsFilter has been published in open access in its entirety, without a stripped-down free version and a paid "for real" one. According to one of the developers, Cloud.ru is deliberately sharing its work with the market, and in the article promises to explain the motivation behind this step.

"I am one of the developers of this tool and would like to tell you more... to answer the question: why would a cloud provider share its work with the market for free?" — one of the

GuardrailsFilter developers, in a blog post on Habr.

The logic for a cloud provider is straightforward: the easier and safer it is for companies to connect LLMs, the more workload they move to the cloud. The open-source tool lowers the entry barrier and, at the same time, serves as a showcase of Cloud.ru's engineering expertise.

What It Means

Anonymizing data before sending it to an LLM is turning from homegrown workarounds into a ready-made, reusable component. For teams that have so far been held back by security requirements, there is now a free way to connect cloud models without handing over confidential information — while also being able to see exactly how the filtering works, right in the source code.

Frequently Asked Questions

What is GuardrailsFilter?

It is a transparent reverse proxy from Cloud.ru placed between the client and the LLM provider. It removes sensitive data from requests to the model and restores it in the responses, without requiring the integration code with the application to be rewritten.

Is GuardrailsFilter paid?

No. Cloud.ru published the source code in open access in July 2026; the tool is distributed free of charge as open source and works with any LLM providers.